The volume of data from medical imaging is growing and consumes high costs of digital data storage. A method that can reduce the image size, fast retrieving and preserving the critical medical details of the image is a highlight in the research. Therefore, fractal image compression (FIC) is essential to obtain a substantially higher compression ratio (C ratio) without perceptible image degradation, which is clinically essential for effective diagnostic performance. The encoding phase in full-search FIC is time-intensive as a sequential search must be performed through a massive domain pool to find the best-matched domain for each block of ranges. This study proposes an improved FIC method based on hierarchical clustered peer adjacent domain and unsupervised multiclass support vector machine (SVM) mapping for computed radiography (CR). The optimal fractal parameters were proposed to increase FIC encoding efficiency using this approach. Combining the peer-adjacent domain with the Pearson correlation coefficient (PCC) is designed to reduce computations, reducing the number of domain blocks in a domain pool. The PCC being used for domain block classification based on correlation value speeds up the encoding. The unsupervised multiclass SVM and K-means clustering are developed in the study to improve the mapping process. The novelty of the proposed approach lies in the use of hierarchical clustered peer adjacent domains with multiclass SVM for accurate domain-range mapping, resulting in a high compression ratio with reduced storage, fast retrieving time, and high reconstructed image quality. The proposed method was tested using various standard test image and two sets database of medical images from The Cancer Imaging Archive (TCIA) and the Japanese Society of Radiological Technology (JSRT). The performance of the proposed optimal fractal parameters is evaluated using the Society of Motion Picture and Television Engineers (SMPTE) image and computed radiography lung images. The results show that the optimal factual parameters for increasing encoding efficiency are quadtree threshold (QTH) equal to 0.2, range size is (4,8), and three decoding iterations. The proposed method shows good performance in the encoding time reduction evaluation for the standard test image in terms of peak signal-to-noise ratio (PSNR), compression time, and compression ratio, with 27.27 dB, 6.88 s, and 16.13, respectively. Eva;uation of various medical image modalities and sizes from TCIA images demonstrates that the proposed method can compress larger images better than small images. For the 16.3 MB 4096 x 4096 mammography image, the proposed method retrieves the compressed image less than a minute with 39.5 dB. The implementation of multiclass SVM and K-means clustering has further improved the compressed image quality. The results for the proposed method evaluation executed using 360 CR images with and without chest lung nodules showed the high quality of reconstructed images with a PSNR equal to 41 dB for a range size of (4,8) and encoded less than a minute. The proposed method saved the storage about 95.6 percent with stored only 358 kB out of 8193 kB original size image. The finding demonstrates that the proposed method obtained high quality reconstructed images with more extensive storage and adequate encoding time.
